Abstract
BACKGROUND: The history from a relative or caregiver is an important tool for differentiating neurodegenerative disease. We characterized patterns of caregiver questionnaire responses, at diagnosis and follow-up, on the Cambridge Behavioural Inventory (CBI). METHODS: Data-driven multivariate analysis (n = 4952 questionnaires) was undertaken for participants (n = 2481) with Alzheimer's disease (typical/amnestic n = 543, language n = 50, and posterior cortical n = 50 presentations), Parkinson's disease (n = 740), dementia with Lewy bodies (n = 55), multiple system atrophy (n = 55), progressive supranuclear palsy (n = 422), corticobasal syndrome (n = 176), behavioral variant frontotemporal dementia (n = 218), semantic (n = 125) and non-fluent variant progressive aphasia (n = 88), and motor neuron disease (n = 12). RESULTS: Item-level support vector machine learning gave high diagnostic accuracy between diseases (area under the curve mean 0.83), despite transdiagnostic changes in memory, behavior, and everyday function. There was progression in CBI subscores over time, which varied by diagnosis. DISCUSSION: Our results highlight the differential diagnostic information for a wide range of neurodegenerative diseases contained in a simple, structured collateral history. HIGHLIGHTS: We analyzed 4952 questionnaires from caregivers of 2481 participants with neurodegenerative disease. Behavioral and neuropsychiatric manifestations of neurodegenerative disease had overlapping diagnostic boundaries. Simple questionnaire response patterns were sufficient for accurate diagnosis of each disease. We reinforce the value of a collateral history to support a diagnosis of dementia. The Cambridge Behavioural Inventory is sensitive to change over time and suitable as an outcome measure in clinical trials.